Title: Towards Performance Guarantees in Emerging Wireless Network Applications

Abstract: The growing interest in mobile systems and Internet-of-Things (IoT) has engendered complex technical challenges ranging from efficient utilization of radio spectrum to developing applications on platforms widely varying in computational power and connectivity. Scalability is increasingly emphasized with exploding number of connected devices, complexity of applications and network data demands with proportionate pressure on limited radio spectrum resources. Our work picks two specific problem domains and explores algorithms that provide performance bounds while scaling to large problem instances. The domains we target are related to distributed radio spectrum monitoring and computation offloading.

In distributed spectrum monitoring we target the spectrum patrolling problem where unauthorized transmitters are detected using a distributed set of spectrum sensors. We specifically consider a crowdsourcing model where a large number of inexpensive sensors are deployed to monitor the radio spectrum. We address different versions of transmitter detection and localization problems, specifically considering the limited budget for the sensors. We develop algorithms to reduce the cost of running a crowdsourced spectrum monitoring system and improve the accuracy of transmitter detection and localization. We also develop FPGA-based spectrum sensors and benchmark the performance improvements in terms of lower latency and energy consumptions than conventionally used sensors that use commodity embedded processor boards. Our future work in this space considers the application of such distributed spectrum monitoring for identifying GPS spoofing devices in the wild.

Our second problem domain is related to computation offloading from weakly powered devices to more powerful cloud servers. Because of the uncertainty inherent in wireless connectivity, and the presence of a large number of devices, deciding which part of the computation to offload or where to offload is challenging. To address this, we propose algorithms that optimize the process of offloading. We focus on providing probabilistic guarantees on the performance of offloaded applications in the presence of channel errors. We further suggest a technique to minimize the completion time of the offloaded application using a novel scheduling technique called task duplication. We show the effectiveness of our algorithm via trace-driven simulation.